[Techtalk] ping -s and mtu size

[Rudy Zijlstra]

Carla Schroder wrote:
> hey all,
>
> Is the ping -s command something that is useful in the real world? The only 
> times I've heard of MTU size problems was when a game dev thought he could 
> use 10,000 byte packets or something equally nuts, and of course TCP broke 
> them up into smaller fragments, which horked game performance.
>
> Suppose you're running ping -s and you carefully increase the size until you 
> start seeing dropped packets. Then what? how do you figure out what is 
> causing the problem?
>
> Carla
>   
Hi Carla,

i am using "ping -s -i " as a poor mans solution to get a feeling for 
link quality between two points. For example i have a bridged link over 
two 11g access points. That means theoretical throughput is at link 
level 54Mbps if the antenna's are well positioned (its a link over about 
300m). In practice an 11g link will have a performance on TCP level of 
about 20Mbps. I've used ping -s -i to have a feeling for how much data i 
could push through (and thus how well the antenna's were positioned).

Another usage is for EMC measurements. In such cases we need traffic 
passing through the link at representative data rates. ping -s is a 
simple and cheap way of generating the needed traffic, and fragmentation 
is actually helping in such a case.

third scenario: test ICMP implementation in a TCP/IP stack. I've crashed 
TCP stacks using ping -s -i....

With respect to your dropped packets, that is something you can only 
figure out when controlling the complete path, so both end-points and 
the complete transmission path in between. I have no idea how to give a 
general answer to that one. Too much environmentally dependent.

Cheers,

Rudy